• In this and the next two exercises, we find some approximate
    eigenvectors, in the sense of Corollary 8.1.29 (page 493).
    For example, let the  3x3  matrix  B  be defined as in the 9-th
    homework set, and let the sequence  G(n)  be as defined there,
    with the starting values  G(0) = G(1) = G(2) = 1.  For a given  m,
    say  m = 10,  let  x  be the the transpose of  (G(m),G(m+1),G(m+2)).
    what are the best values of  alpha  and  beta  satisfying the
    hypothesis of 8.1.29, namely
    
                   alpha x  <=  B x  <=  beta x ?
    
    (By best, one means the greatest  alpha  and the smallest  beta.)
    Hence what estimate does 8.1.29 provide for  rho(B) ?
    
    
  • Let  Dn  be the  n x n  matrix
    
                       0 1 0 0 ... 0 0 0
                       1 0 1 0 ... 0 0 0
                       0 1 0 1 ... 0 0 0
                             ...
                       0 0 0 0 ... 1 0 1
                       0 0 0 0 ... 0 1 0
    
    Using the transpose of  (5,8,10,10, ... ,10,10,8,5))  as an
    approximate eigenvector for  Dn,  show that  rho(Dn)  lies
    between  1.6  and  2.0  (for  n > 3).
    
    Optional. Verify that the transpose of
    
               (sin pi/(n+1), sin 2pi/(n+1), ... , sin n pi/(n+1))
    
    is a positive eigenvector of  Dn.  What is its eigenvalue? Hence,
    what is the precise value of  rho(Dn)?  Then verify that, as  n
    goes to infinity, the limit of  rho(Dn)  is  2.0.  (You may have
    to brush up your trigonometry!)
    
  • Let  C  be the  3x3 matrix
    
                            1 1 1
                            1 2 1
                            1 1 1
    
    and let  x  be the transpose of  (12,17,12). What are the best
    alpha and beta for which one may invoke  8.1.29?  What bounds does
    one therefore obtain on the top eigenvalue  rho(C)?
    
    Next, find another approximate eigenvector, with integer entries,
    which yields a sharper estimate of the eigenvalue  rho(C).
    
    Finally, find the precise value of  rho(C),  and an associated
    eigenvector. (If you can _guess_ an eigenvector, from what has
    come before, so much the better!)
    
  • This exercise may help elucidate the power method, which was
    introduced in the somewhat mysterious Exercise 7 on page 63,
    which we had for February 16. Here we are able to carry out
    the power method for  C,  and moreover analyze the situation
    with the methods of Section 8.2.
    
    Let us continue with the matrix  C  of the previous exercise.
    According to Theorem 8.2.8 (page 499), the powers of  C/rho(C)
    approach a very simple matrix  L.  All columns of  L  are multiples
    of a single vector  x,  where  x  is an eigenvector with eigenvalue
    rho(C). Moreover, according to clause  (j)  on page 498, the
    convergence with respect to the infinity-norm is of the order
    of  K r^n,  where  r  is any number larger than the quotient of
    the next-to-top eigenvalue to the top eigenvalue, and  K  is a
    constant.
    
    What is this ratio for the matrix  C?  What is the eighth power
    of this ratio? Compute the eighth power of  C.  (Don't worry about
    the scalar factor that should really be there in a fastidious
    invocation of Theorem 8.2.9.) Let  c  be the first column of the
    eighth power of  C.  How close is  c  to being an eigenvector of  C?
    (You could answer this question by again obtaining best  alpha  and
    beta  for an application of  8.1.29.  One expects  alpha  and  beta
    to agree within roughly  alpha*r^8.)
    
  • Optional. Carry out a similar power analysis of the matrix  B  from
    the 9-th Exercise Set, including the rate of convergence. You should
    already have found (above) an approximate value of  rho(B),  but for
    this exercise will also require an approximate value of the next-to-
    largest eigenvalue. This may be obtained either by deflation (Ex 8,
    page 63), or simply by dividing the characteristic polynomial by
    (x - r),  where  r  is the approximate value of  rho(B).